Burcu KIRAN, PhD Department of Econometrics Istanbul University Turkey E-mail: kburcu@istanbul.edu.tr THE EXPECTATĐON HYPOTHESĐS OF THE TERM STRUCTURE OF INTEREST RATES: EVĐDENCE FROM SELECTED HĐGH INCOME OECD COUNTRĐES Abstract.This paper empirically investigates the expectation hypothesis of the term structure of interest rates for selected high income OECD countries over the period from 1990:01 through 2010:04 by means of fractional cointegration approach. The results show that long term and short term interest rates for all selected countries are not fractionally cointegrated, implying unvalidity of the expectation hypothesis of the term structure. Keywords:Expectation Hypothesis, Term Structure of Interest Rates, Fractional Cointegration. JEL Classification C2, E43.Introduction The expectation hypothesis of the term structure suggests that a long term interest rate can be represented as a weighted average of the present and expected future short term interest rates. In other words, according to this hypothesis, long term interest rates should reflect future short term changes. Since this simple theory has an importance in assessing the impact of monetary policy of a country, predicting interest rates, exchange rates, economic activity and providing information about expectations of participants in financial markets, an enormous amount of attention has been given to investigate the expectation hypothesis of the term structure of interest rates in financial economics literature. The expectation theory of the term structure is the joint hypothesis that agents hold national expectations and term premia are invariant. These time varying term premia or forecast errors which appear biased when viewed ex post, cause to reject the expectation hypothesis of term structure (Hezaji et. al., 2000). Different studies which use different econometric methods, test different implications of the

Burcu Kıran ____________________________________________________________expectations theory and look at different interest rate maturities give contradictory results. The studies including Shiller(1979), Fama(1984), Mankiw(1986), Campbell and Shiller(1987, 1991), Boothe(1991), Campbell(1995), Siklos and Wohar(1996), Sarno et. al(2007) reject the expectation hypothesis of term structure while the studies including Hall et. al(1992), Engsted and Tanggaard(1994), Hardouvelis(1994), Hurn et. al(1995), Cuthbertson(1996) support the expectation hypothesis. Generally, the cointegration methodology is used in order to test expectation hypothesis of term structure in the literature. Existence of a long run relationship between long term and short term interest rates is a necessary condition for corresponding hypothesis to hold(Campbell and Shiller, 1987). If the expectation hypothesis holds, the term spread is stationary, hence, short term and long term interest rates are cointegrated. Since the traditional cointegration methods are too restrictive because of they assume that all the variables are integrated of order one, (1)and restrict error correction term to be(0), it is possible to use other recently developed techniques for expectation hypothesis. The aim of this paper is to investigate the expectation hypothesis of the term structure of interest rates for selected high income OECD countries over the period from 1990:01 to 2010:04 by using fractional cointegration concept which requires only a mean reverting relationship between the considered series. Lardic and Mignon(2004) test the expectation hypothesis by investigating the fractional cointegration relationship between short term and long term interest rates for G7 countries and find that the interest rates are fractionally cointegrated for the countries except Germany. The difference of our paper from the paper of Lardic and Mignon(2004) is that we use fractional cointegration approach introduced by Gil-Alana(2003) and Caporale and Gil-Alana(2004). The outline of the paper is organized as follows: Section 2 briefly discusses the expectation hypothesis of the term structure of interest rates. Section 3 outlines the econometric methodology. Section 4 discusses the data and reports the empirical results. Finally, Section 5 concludes. 2. The Expectation Hypothesis of the Term Structure of Interest Rates The expectation theory of the term structure of interest rates is a relationship (n) between a longer termnand a shorter term-period interest rate m-period t (m) (n) interest rate . In details, withn>mthe weighted average of the, is t t (m) expected futurem, and plus a term premium. Theinterest rate period t specification of this relationship can be seen in below:

The Expectation Hypothesis of the Term Structure of Interest Rates: Evidence ..... __________________________________________________________________ k−1 1 (n) (m) ∑t+i =E R+Ek=n m t t t t ,(2.1)ki=0 where is the expectation operator conditional on information at timet and t t gives the term premium which is a predictable excess return on then-period bond over them-period bond. This term premium may vary withn andm but it is assumed to be constant through time (Campbell and Shiller, 1991). By re-arranging Equation (2.1), the yield spread between then-period rate and the m-period rate can be obtained as follows: k−1i 1 (n−m) (n) (m) (m) t t t∑∑tt+it t S=R−R=EΔR+E (2.2) ki=1k=1 (n) (m) (n) (m) Here, if and have a unit root, interest rate spread (−R) will be a t t t t (n) (m) stationary process. Thus, it can be said that and have a cointegrating t t relationship in the long run with a cointegrating vector (-1, 1)’ and expectation hypothesis of term structure holds (Lardic and Mignon, 2004). On the other hand, there is a possibility that the error term in the cointegrating regression might be (n) (m) fractionally integrated, rather than stationary. In other words, if−Ra is t t (n) (m) long memory process, then and are said to be fractionally cointegrated. t t (n) (m) In this context, deviations from the long run relationship shared by and t t take a long time to dissipate and return these two series to their equilibrium relationship (Bekdache and Baum 2000). 3. Methodology In this paper, we investigate the expectation hypothesis of the term structure of interest rates by means of fractional cointegration approach, since traditional cointegration methods have low power when the residuals are mean reverting but not(0). This approach allows residuals to be fractionally integrated rather than stationary. For this purpose, we follow fractional cointegration concept of Gil-Alana(2003) and Caporale and Gil-Alana(2004) based on Robinson(1994a) test. In order to be able to understand the theoretical structure of their concept, it is better to give a brief description of Robinson test. Robinson(1994a) considers the following regression model, ' yβz+x ,t1, 2,.... (3.1) t t t

Burcu Kıran ____________________________________________________________' whereyis the observed time series fort1, 2,...T,=(β,...,β)is a(k1)t1k vector of unknown parameters, is a(k1) vector of deterministic regressors t such as an intercept or a linear trend. And the regression errors can be explained t as follows: d (1 )x=u ,t1, 2,.... (3.2) t t where is the lag operator anduis an(0)process. Here,dcan take any real t value. Robinson suggests a Lagrange Multiplier (LM) test statistic for testing unit roots and other forms of nonstationary hypotheses, embedded in fractional alternatives. The main advantage of the procedure is that it tests unit and fractional roots with a standard null limit distribution, which is unaffected by inclusion or not of deterministic trends. The notation of the LM test statistic under the null hypothesis:d dcan be seen in below: 0 0 1 2 T1 2 ˆ rˆ=A aˆ(3.3) 2 σˆ whereTis the sample size and T−1T−1 −2π−2 22π1 1− aˆ=(λ)g(λ;τˆ)I(λ);ˆ=σ(τˆ)=g(λ;τˆ)I(λ); ∑j j j∑j j Tj=1Tj=1 −1 T−1T−1T−1T−1 2 ˆ 2 ' ' A= (λ)−(λ) ˆε(λ)×ˆε(λ) ˆε(λ)×ˆε(λ) (λ)∑ ∑ ∑ ∑ j j jj jj j T j=1j=1j=1j=1 j2 ψ(λ)=log 2 sin;ˆ(λ)=logg(λ;τˆ );λ=; jj jj j 2∂τT 2 ˆ=arg min*σ(τ). τ∈T * Here,I( )the periodogram of is u andTa compact subset or the is jt Euclidean space. Robinson(1994a) showed that the test statistic under certain regularity conditions is as below: rˆ→N(0,1)asT→ ∞. (3.4) d Thus, a one sided100 %level test of the null hypothesis:d dagainst the 0 0 alternative:d>d“Reject ifis given by the rule rˆ>z”. Conversely, a 1 0 0

The Expectation Hypothesis of the Term Structure of Interest Rates: Evidence ..... __________________________________________________________________ one sided100 % level test of:d dagainst the alternative:d d is 0 0 1 0 given by the rule “Reject ifrˆ−z”. Following these rules, Gil-Alana(2003) 0 and Caporale and Gil-Alana(2004) suggest a fractional cointegration concept based on the following model: d+θ (1 )e=v ,t1, 2,.... (3.5) t t whereethe OLS residuals from the cointegrating regression and is v is(0). t t The nullH:θ0 hypothesis is tested against the one sided alternative 0 H:θ0. If hypothesis on the estimated residuals is rejected, there is an 1 0 evidence of fractional cointegration of a certain degree since the residuals are integrated of a smaller order than the individual series. If we cannot reject the null hypothesis, it can be concluded that there is no evidence of fractional cointegration since the integration order of the residuals are same as the univariate series.4. Data and Empirical Results In order to examine the expectation hypothesis of the term structure of interest rates for selected high income OECD countries: Belgium, France, Italy, Spain, Canada, Switzerland and UK, we use quarterly series of short term and long term interest rates over the period from 1990:01 through 2010:04. We consider 3-month treasury bill rates (TR) for short term interest rates and 10-year government bond rates (GOV) for long term interest rates. The source of the data is IMF’s International Financial Statistics (IFS) database. The first step of the empirical analysis is to investigate the integration order of the individual series. For this purpose, we perform Augmented Dickey Fuller (ADF), Philips and Perron (PP) and Kwiatkowski-Phillips-Schmidt and Shin (KPSS) unit root tests. These tests differ in the null hypothesis: The null hypothesis of the ADF and PP tests is that a time series contains a unit root, I(1) process, while the KPSS test has the null hypothesis of stationarity, I(0) process. The results of unit root tests under the different null hypothesis are characterised by four possible outcomes (Barkoulas et. al, 1997): i) When we reject the null hypothesis of the ADF and PP tests and we cannot reject the null hypothesis of the KPSS test, a time series is stationary. ii) Conversely, failure to reject a unit root by ADF and PP tests and rejection stationarity by KPSS test supports that a time series is nonstationary. iii) Failure to reject a unit root and stationary null hypotheses shows that the series are not sufficiently informative with respect to the low frequency properties. iv) rejection of null hypotheses indicates that series are not well represented as either I(0) and I(1), which indicates that the series appear to be a long memory process. The results of the ADF, PP and KPSS unit root tests are reported in Table 1.

Burcu Kıran ____________________________________________________________Table 1: The results of ADF, PP and KPSS unit root tests Countries Variables ADF PP KPSS a GOV -2.484 -2.222 0.245 a a ΔGOV -6.297 -6.243 0.191 Belgiumb TR -1.967 -2.138 0.200 a a ΔTR -7.471 -7.478 0.153 a GOV -3.051 -2.433 0.232 a a ΔGOV -5.998 -5.983 0.122 Franceb TR -2.016 -2.293 0.210 a a ΔTR -7.956 -7.951 0.103 a GOV -1.108 -1.688 0.263 a a ΔGOV -6.263 -5.357 0.161 Italya TR -2.419 -2.059 0.238 a a ΔTR -6.011 -5.986 0.085 a GOV -2.353 -1.700 0.277 a a ΔGOV -5.164 -5.284 0.030 Spaina TR -2.725 -1.768 0.257 a a ΔTR -5.092 -5.090 0.160 b GOV -2.403 -2.400 0.199 a a ΔGOV -8.135 -8.130 0.105 Canadaca b TR -5.052 -3.190 0.124 ΔTR - - 0.076 c b GOV -3.315 -2.763 0.201 a ΔGOV - -6.493 0.062 Switzerlanda TR -2.126 -2.054 0.226 a a ΔTR -6.311 -6.358 0.061 b a GOV -3.713 -2.418 0.238 a ΔGOV - -7.076 0.166 UK b a TR -3.673 -2.874 0.906 a ΔTR - -5.270 0.188 a, b c and denote that the unit root null hypothesis is rejected at the 1%, 5% and 10% significance levels. As can be seen from the table, the results of ADF test indicate that long term interest rate series are nonstationary for the countries except Switzerland and UK while short term interest rates are nonstationary for the countries except Canada and UK. According to PP unit root test results, long term interest rates are nonstationary in level but stationary in the first difference for all selected countries while short term interest rate series are nonstationary for the countries except Canada. On the other hand, KPSS test results indicate that both long term and short term interest rates are nonstationary in level for all countries. These contradictory results may arise because the(d)concept whered is an integer, is too restrictive. Next, we also perfom Robinson(1994a) univariate unit root test on the individual series. The one sidedrˆstatistic values ford1are reported in Table 2.

The Expectation Hypothesis of the Term Structure of Interest Rates: Evidence ..... __________________________________________________________________ Table 2: The results of Robinson test for unit root Countries Variables Test statistics b GOV 0.084 Belgiumb TR -0.095 b GOV 0.250 France b TR -0.143 b GOV 0.513 Italyb TR 0.610 b GOV 0.686 Spainb TR 1.517 b GOV 0.159 Canada TR 0.579b b GOV 0.413 Switzerlandb TR 0.209 b GOV 0.217 UK b TR 0.464 b indicates nonrejection values of the unit root null hypothesis (d1) at the 95% significance level. 0 In bold: the absolute value of the minimum of the Robinson test statistic. We consider only the test whereuis assumed to be white noise, for the specification with an intercept. t The results show that there is evidence of a unit root at the 95% significance level for short term and long term interest rate series of all countries. Having found that all series exhibit a unit root behavior, following Gil-Alana(2003) and Caporale and Gil-Alana(2004), we investigate whether there is a fractional cointegration relationship between short term interest rates and long term interest rates. The long term interest rates are regressed against short term interest rates for each country, consistently with expectation hypothesis theory of term structure, and the residuals are obtained from these cointegrating regressions. The results of Robinson test applied on the residuals are tabulated in Table 3. Here, different values ofdare considered, thus testing for a unit root (d1) but also other fractional possibilities. Table 3: The results of Robinson test on the estimated residuals dBelgium France Italy Spain Canada Switzerland UK 0 0.009.843 8.852 9.400 10.323 12.191 8.325 12.885 0.059.052 10.015 11.684 9.420 8.386 7.718 12.440 0.108.683 9.681 11.155 8.989 7.912 7.111 11.960 0.156.504 11.4448.549 7.428 8.290 9.317 10.604 0.205.898 10.8938.101 6.937 7.876 8.924 10.033 0.2510.3099.446 5.296 7.442 8.502 7.645 6.440

Burcu Kıran ____________________________________________________________0.307.182 5.939 8.846 4.701 6.991 8.050 9.694 0.356.526 7.572 6.714 5.437 9.0538.237 4.118 0.408.3926.241 4.935 6.051 7.069 7.623 3.549 0.457.009 2.998 7.7185.766 4.437 5.572 6.547 0.505.290 3.945 5.092 6.009 6.398 2.468 7.036 0.554.616 5.461 4.817 3.462 6.3565.794 1.961 b 0.604.149 4.908 4.347 2.990 5.201 1.479 5.684 b 0.653.884 2.531 3.694 4.356 4.621 1.023 5.027 b 0.704.057 0.593 4.3913.255 3.811 3.430 2.087 b 0.753.512 0.190 3.7822.834 3.278 2.985 1.659 bb 0.802.553 1.249 2.432 2.760 2.988-0.1873.204 b b 0.852.134 0.858 2.052 -0.538 2.6582.262 2.486 b b 0.901.729 0.485 1.692 1.787 2.008 -0.864 2.146 b b b b b b 0.951.341 0.332 1.554 -1.166 1.6691.355 1.336 b b b b bbb 1.00 0.968 -0.202 1.037 0.911 1.124-1.4451.226 b indicates nonrejection values of the null hypothesis at the 95% significance level.In bold: the absolute value of the minimum of the Robinson test statistic. We consider only the test whereuis t assumed to be white noise, for the specification with an intercept. The results show that nonrejection values occur at the values ofd0.95 and 1 for Belgium, Italy, Spain and Canada; dand 10.60, 0.65, 0.70, 0.75, 0.80, 0.85, 0.90, 0.95 for Switzerland; d0.80, 0.85, 0.90, 0.95 and 1for France andd1for UK. It is clear that unit root null hypothesis (d1) cannot be rejected at all. On the other hand, for all countries except Switzerland, the minimum of the absolute values of the Robinson test statistics, which is indicated in bold, corresponds to the value of thed parameter equal to 1. For Switzerland, it can be seen that the unit root null hypothesis cannot be rejected although the minimum of the absolute values of the test statistics occurs at the value ofd0.80. These results indicate that there is no evidence of fractional cointegration relationship between short term and long term interest rates, implying unvalidity of the expectation hypothesis of the term structure for considered countries. Conclusions In this paper, we examine the expectation hypothesis of the term structure of interest rates for selected high income OECD countries: Belgium, France, Italy, Spain, Canada, Switzerland and UK, over the period from 1990:01 through 2010:04 in the context of fractional cointegration. As a first step, the stationarity properties of the short term and long term interest rate series are investigated by using ADF, PP and KPSS unit root tests. Since these traditional unit root tests give contradictory results, we also apply Robinson(1994a) test on the individual series for a unit root case (d1). According to the obtained findings, the evidence of a

The Expectation Hypothesis of the Term Structure of Interest Rates: Evidence ..... __________________________________________________________________ unit root for short term and long term interest rate series is found for all countries. In the next step, we investigate whether there is a fractional cointegration relationship between short term and long term interest rates following fractional cointegration procedure of Gil-Alana(2003) and Caporale and Gil-Alana(2004). The results indicate that there is no evidence of fractional cointegration relationship between short term interest rates and long term interest rates. Hence, it can be concluded that the expectation hypothesis of the term structure is not valid for considered coutries. REFERENCES [1]Barkoulas, J.W., Labys, C., Onochie, J. (1997),Fractional Dynamics in International Commodity Prices.Journal of Futures Markets, 17, 161-189; [2]Bekdache, B., Baum, C. (2000),A Re-evaluation of Empirical Tests of the Fisher Hypothesis.Working Paper, Department of Economics, Wayne State University, 1–24; [3]Boothe, P. (1991),Interest Parity, Cointegration, and the Term Structure in Canada and the United States.Canadian Journal of Economics, 24, 595–603; [4] Campbell, J.Y. (1995). “Some Lessons from the Yield Curve”,Journal of Economic Perspectives, 9, 129-152. [5]Campbell, J.Y., Shiller, R.J. (1987),Cointegration and Tests of Present Value Models.The Journal ofPolitical Economy, 95, 1062-1088; [6]Campbell, J.Y., Shiller, R.J. (1991),Yield Spreads and Interest Rate Movements: A Birds Eye View.Review of Economic Studies, 58, 495–514; [7]Caporale, G.M., Gil-Alana, L.A. (2004),Fractional Cointegration and Tests of Present Value Models.Review of Financial Economics, 13, 245-258; [8]Cuthbertson, K. (1996),The Expectations Hypothesis of the Term Structure: The UK Interbank Market.Economic Journal, 106, 578–592; [9]Engsted, T., Tanggaard, C. (1994),A Cointegration Analysis of Danish Zero-Coupon Bond Yields;Applications of Finance and Economics,4, 265 –278; [10]Fama, E.F. (1984),The Information in the Term Structure.Journal of Financial Economics, 13, 509-528; [11]Gil-Alana, L.A. (2003),Fractional Cointegration in Macroeconomic Time Series.Oxford Bulletin of Economics and Statistics, 65, 517-524; [12]Hall, A.D., Anderson, H.M., Granger, C.W.J. (1992),A Cointegration Analysis of Treasury Bill Yields .Review of Economics and Statistics, 74, 116-126; [13](1994),Hardouvelis, G.A. The Term Structure Spread and Future Changes in Long and Short Rates in the G7 Countries.Journal of Monetary Economics, 33, 255–283; [14]Hejazi, W., Lai, H., Yang, X. (2000),The Expectations Hypothesis, Term Premia and the Canadian Term Structure of Interest Rates; TheCanadian Journal of Economics, 33, 133-148;

Burcu Kıran ____________________________________________________________[15]Hurn, A.S., Moody, T., Muscatelli, V.A. (1995),The Term Structure of Interest Rates in the London Interbank Market.Oxford Economic Papers, 47, 418–436; [16]Lardic, S., Mignon, V. (2004),Fractional Cointegration and the Term Structure.Empirical Economics, 29, 723-736; [17]Mankiw, N.G. (1986),The Term Structure of Interest Rates Revisited. Brookings Papers on Economic Activity61–96;, 1, [18]Robinson, P.M. (1994a),Efficient Tests of Nonstationary Hypothesis. Journal of the American Statistical Association, 89, 1420-1437; [19]Sarno, L., Thornton, D.L., Valente, G. (2007),The Empirical Failure of the Expectation Hypothesis of the Term structure of Bond Yields.Journal of Financial and Quantitative Analysis, 42, 81-100; [20]Shiller, R.J. (1979),The Volatility of Long-Term Interest Rates and Expectations Theories of the Term Structure.Journal of Political Economy, 87, 1190–1219; [21]Siklos, P. L., Wohar, M. E. (1996),Cointegration and the Term Structure: A Multicountry Comparison.International Review of Economics and Finance, 5, 21-34.